More detail: Move your organization from class to work
Optional reading for this lesson. Everything you must do or decide is in the lesson.
This page goes with the Week 4 lesson Move your organization from class to work (course/workforce/week4/lessons/anw-w4-l12.md). You don't need it to finish the lesson. Read it when you want the longer version.
The method on one page
| Move | Where you learned it this week |
|---|---|
| Write what you expect, before the work runs | "Read your team's work and give each one a thumbs up or down" |
| Grade it at two levels, against your rules | "Turn what you found into the rules for this team" |
| Let Evy grade, and check by hand as much as your step says | "Compare your grades with Evy's and teach Evy" |
| Fix one problem at a time: you approve the plan, the agents make the change and rerun the whole golden set | "Make the fix, rerun the golden set, and keep or undo it" |
| Save the great as examples and the mistakes as tests | "Save your golden examples, and turn mistakes into regression tests" |
| Read how your teams did each morning | "Turn on Evy's daily check" |
The handoff notes
In "Move your organization from class to work" your Chief writes the first handoff note for a team, in a dated folder under runs/: what's in progress, what's waiting on you, and anything odd from the last job. Ask for a new one after each job you care about, and start the next request with "Read the newest handoff note for this team first."
What everyday mode changes
Your records are marked as everyday work instead of course practice. Everything you set up keeps running: Evy's daily check, the Charter Steward's daily check, your rule sheets, Kansa's watch lists, your shelves, and the standing rule for anything new. The latest ruling wins, so you can switch back any time.
Evy's time
You can pick the time Evy's daily check runs by asking your Chief of Staff. Your morning digest, a short brief your app writes on a timer, starts Evy's check first at that time. Your first brief of the day still runs it if it hasn't run yet, so a brief before your time runs it early, and it never runs twice in a day. We've tested the digest's instructions and the timing. We haven't yet tested the app's own timer (Claude Code's Routines, Codex's Scheduled list), so your first morning is the test, and your Chief checks it with you.
While your computer is off
Your agents run on your computer, only while it's awake and the app is open. Your first brief each day catches up: it runs once what only needs the latest run, replays each missed day where every run counts, and skips what's too late, saying so.
The always-on options, in more detail
| Option | Good for | Watch out for |
|---|---|---|
| Cloud routines in Claude Code | Read-only nightly jobs, like a summary of yesterday | They run from your copy on GitHub, so they see only what you saved there. They use your connected tools without asking, so connect read-only tools. They can't read your chats with your Chief. |
| ChatGPT's cloud tasks | Jobs that need only your connected tools and uploads | They can't reach your workbench folder. |
| An always-on computer | Everything, all the time | The hardest path: a small server that never sleeps, private network access, and every tool wired in again. |
The course doesn't teach these. Try them when you're ready, one job at a time, and keep wiring anything new into Langfuse.
The Thinking Partner (Riff) and the Research Planner (CLEAR), in more detail
Riff. Riff reads your founder brief and priorities, so its two voices fit what you said matters to you. For a big or values-heavy choice it starts with one line on the general principle, then applies it to yours. You can ask for a gentler or sharper riff, or for only one voice. If a choice needs a doctor, a lawyer or a financial adviser, Riff says so and names the kind of person who can help. When you talk to Riff directly it saves nothing; to keep a riff, ask your Chief to open a job and have Riff write it there.
CLEAR. CLEAR reads what you've already written down before it asks you anything, so it never asks what it could have read. Its brief has seven parts, always in the same order: the objective, the context and limits, the ranked questions, the sources to prefer, how sure the answers need to be, how the researcher reports, and what "done" means. Your Chief runs a checker on the brief, and if it finds a problem, CLEAR fixes exactly that in a new version. When the External Researcher (Scratch) brings findings back, the same checker lists each question as answered, a stated gap, or missing. A question with a clear gap and a reason counts as covered. If the findings miss a question, CLEAR asks for one rerun. A second miss means the brief was wrong, and CLEAR says what to change. CLEAR can also frame a vague build idea or a decision the same way.